Arrakis / Models / gpt-oss-120b
gpt-oss-120b GPU requirements
gpt-oss-120b has 117B parameters. To serve it, plan for about 280 GB of GPU memory in 16-bit, 140 GB in 8-bit or 70 GB in 4-bit, including about 20% for the KV cache and activations.
It is a mixture of experts: 128 experts, 4 of them active for each token. All of them stay in GPU memory, so plan memory for all 117B parameters.
What it takes to serve it
GPU memory to run it for inference, with the smallest GPU setup that holds it and the lowest price on Arrakis right now.
| Precision | Memory and the smallest setup |
|---|---|
| 16-bit | 280 GB: 2 × H200, from $8.06 per hour for all 2 |
| 8-bit | 140 GB: one H200, from $4.03 per GPU-hour |
| 4-bit | 70 GB: one H100 or A100, H100 from $1.65 per GPU-hour |
What it takes to fine-tune it
QLoRA trains adapters on a 4-bit frozen model, LoRA on a 16-bit one, and full fine-tuning updates every weight. Figures are before activations, which grow with batch size and sequence length.
| Method | Memory and the smallest setup |
|---|---|
| QLoRA | 82 GB: one RTX PRO 6000, from $1.90 per GPU-hour |
| LoRA | 280 GB: 2 × H200, from $8.06 per hour for all 2 |
| Full fine-tuning | 1,869 GB: 8 × B300, from $36.80 per hour for all 8 |
About the model
From its card and files on Hugging Face. Memory figures here follow the GPU memory guide.
- Parameters
- 117B
- Experts
- 128
- Experts for each token
- 4
- Context window (tokens)
- 131,072
- Published in
- FP8
- License
- apache-2.0
- Released
- 2025-08-04
- Weights
- openai/gpt-oss-120b
Deploy it on Arrakis
Open Models in the console, choose vLLM, SGLang or llama.cpp, and rent a GPU it fits on. You get an OpenAI-compatible endpoint with its own API key, and root SSH to the machine. Or run it yourself on any machine with enough GPU memory:
vllm serve openai/gpt-oss-120b --tensor-parallel-size 2Questions
Live prices for every GPU are on the GPU prices page.